Skip to content

AP Statistics · technique

When to reach for the calculator

On this page · 5 sections

The May 2027 AP Statistics exam has two sections of equal weight. Section I is 42 multiple-choice questions in 90 minutes, each with four choices (A)–(D). Section II is four free-response questions, also in 90 minutes. You get a formula sheet and tables for the normal, t and chi-square distributions, and a graphing calculator with statistics functions is expected in both sections.

This skill is about deciding when the calculator earns its time. It is not a separate scored task. It shows up in your score because of speed on Section I and clear communication on Section II. On multiple choice you get one point per correct answer and lose nothing for a wrong one, so a calculator only helps if it gets you to the right letter faster. Many questions ask what a p-value means, which design is best, or how a change in sample size moves a margin of error. Those need no calculation at all, and reaching for the calculator only burns time.

On free response, readers score what you write on the page, not what your screen showed. A bare line like "normalcdf(0, 445, 454, 6) = 0.067" tells them the answer but not the model. The safe way is to name the distribution, its parameters and the boundary in words or standard notation, then report the calculator's result. The calculator does the arithmetic; you still have to show the statistics.

The steps

  1. 1

    Read the stem and sort the question first

    Before you touch a key, decide what kind of question it is. Ask: does a correct answer need a number I don't have yet? If the stem asks you to interpret a confidence level, name a bias, or pick the right test, it is conceptual, and you answer it from reasoning. If it asks for a probability, a percentile, a test statistic or a regression prediction, it is computational. About 90 minutes for 42 questions gives you roughly two minutes each. Conceptual questions should take under a minute, which banks time for the computational ones.

  2. 2

    Estimate before you calculate

    Make a rough guess so you can eliminate choices and catch typing errors. For a normal model, use the 68–95–99.7 rule: a value 2 standard deviations below the mean has about 2.5% below it. For a margin of error, remember that quadrupling n halves it. For a regression prediction, plug x into ŷ = a + bx in your head with rounded numbers. Often the estimate alone rules out three choices. If it doesn't, you now know roughly what the calculator should say, so a wrong keystroke shows up at once.

  3. 3

    Match the question to the right command

    Learn the short list cold, because hunting through menus costs more time than the arithmetic. Use normalcdf(lower, upper, μ, σ) for an area under a normal curve. Use invNorm(area to the left, μ, σ) for a percentile or cutoff. Use binompdf(n, p, k) for exactly k successes and binomcdf(n, p, k) for k or fewer. Use tcdf and invT for t areas and critical values. Use 1-PropZInt, 2-PropZTest, TInterval, T-Test, 2-SampTTest and χ²-Test for inference. Use 1-Var Stats for summary statistics and LinReg(a+bx) for a least-squares line. These names are TI-83/84 syntax. Other approved calculators, and any calculator built into the digital testing app, use different command names, but you still write the same inputs on the page: the distribution, its parameters and the boundary.

  4. 4

    Translate the words into the command's inputs

    Most calculator errors are translation errors, not typing errors. "At least 4" for a binomial means 1 − binomcdf(n, p, 3), because binomcdf only counts at or below k. "More than 445 grams" means a lower bound of 445 and an upper bound you set very high, like 1E99. A normal model needs the standard deviation, so if the stem gives a variance of 36, you enter 6. For a sample proportion, the standard deviation is √(p(1−p)/n), not p. Say the inputs to yourself before you press ENTER.

  5. 5

    Check the output against your estimate and the context

    Compare the screen to the guess you made in step 2. A probability must be between 0 and 1. A percentile cutoff for the bottom 10% must sit below the mean. A 95% interval must contain the sample statistic at its centre. In 1-Var Stats, take Sx, the sample standard deviation, not σx, unless the data are a whole population. For a two-sample t procedure, leave Pooled set to No. If the output fails one of these checks, find the translation mistake rather than trusting the screen.

  6. 6

    On free response, write the statistics, then the number

    Readers can't see your calculator, so put the model on the page. Name the distribution and its parameters, show the boundary, and shade a quick sketch if it helps. For inference, name the procedure, state the hypotheses or the parameter, check the conditions, and then give the test statistic, the degrees of freedom and the p-value or interval. Calculator syntax is acceptable only when every input is labelled. Then finish with a conclusion in context, because the calculator never writes that part for you.

  7. 7

    Keep time-boxes and move on

    Set a limit of about three minutes for any single multiple-choice question. If you're still entering data after that, pick your best estimate, mark it, and come back at the end. In Section II, budget about 22 minutes per question and don't redo a calculation just to get one more decimal place. Round your final answer to a sensible number of places, but keep full precision in the steps the calculator carries for you.

Worked through

Problem (free-response style). A school transport office records how long 8 randomly chosen students wait for the morning bus, in minutes: 12, 15, 9, 14, 11, 16, 13, 10. (a) Find the mean and standard deviation of the waits. (b) Construct and interpret a 95% confidence interval for the mean wait for all students. (c) The office claims the mean wait is 15 minutes. Does the interval support that claim?

Step 1, sort the question. Part (a) needs numbers from raw data, so the calculator earns its time. Typing 8 values into L1 takes about 20 seconds, which beats summing squares by hand.

Step 2, estimate. The values cluster around 12 or 13, and they range from 9 to 16. A rough standard deviation is about range ÷ 4 ≈ 2, so you expect something near 2 to 3.

Step 3, run 1-Var Stats on L1. The screen shows x̄ = 12.5, Sx ≈ 2.449 and σx ≈ 2.291. These are a sample of 8 students, not every student, so you report Sx. On the page you write: x̄ = 12.5 minutes, s ≈ 2.45 minutes. Both numbers match the estimate, so you move on.

Step 4, choose the procedure for (b). The population standard deviation is unknown and the parameter is a mean, so this is a one-sample t interval, not a z interval. Write the parameter in words: μ = the true mean morning wait for all students at this school.

Step 5, check conditions in writing, because the calculator never checks them. Random: the students were randomly chosen. Independence: 8 is less than 10% of all students at the school, which is safe to assume because the school has more than 80 students. Normal: n = 8 is small, so look at the data. A quick dotplot from 9 to 16 shows no strong skew and no outliers, so a t procedure is reasonable.

Step 6, run TInterval with Stats or Data input and a 95% level. The screen gives about (10.45, 14.55) with df = 7. On the page, show where it comes from: 12.5 ± t*(2.449/√8), with t* = 2.365 for df = 7. That works out to 12.5 ± 2.365(0.866) = 12.5 ± 2.05. Writing this line shows the reader you know the structure, even though the calculator did the arithmetic.

Step 7, interpret. We are 95% confident that the interval from 10.45 to 14.55 minutes captures the true mean morning bus wait for all students at this school.

Step 8, answer (c) with reasoning, not a new calculation. The claimed value of 15 minutes is above the upper end of the interval, at 14.55. So 15 is not a plausible value for the mean, and the interval gives convincing evidence that the true mean wait is not 15 minutes. It suggests the mean is lower. No second test is needed, because the interval already answers the question.

What this example shows: the calculator handled the summary statistics and the interval in under a minute. Every point the reader awards comes from the parts you wrote yourself. Those are the parameter, the conditions, the formula structure, the interpretation and the link from the interval to the claim.

Common mistakes

  • Reading σx instead of Sx from 1-Var Stats. For sample data you need Sx, the sample standard deviation, which divides by n − 1. Using σx makes the standard deviation too small, so your interval comes out too narrow and your test statistic too big.

  • Using binomcdf(n, p, k) for "at least k". The command gives P(X ≤ k), so P(X ≥ k) is 1 − binomcdf(n, p, k − 1). Entering k instead of k − 1 drops the probability of exactly k, which is often the exact gap between two answer choices.

  • Entering a variance where the command wants a standard deviation. normalcdf and invNorm take σ, so a stem that gives a variance of 36 needs an input of 6. For a sample proportion, use √(p(1−p)/n) as the standard deviation, not p itself.

  • Picking a z procedure when σ is unknown, or setting Pooled to Yes in 2-SampTTest. Means with an estimated standard deviation need t, and AP Statistics expects the unpooled two-sample t. Either slip gives a test statistic and p-value that the key won't match.

  • Writing bare calculator syntax on free response, such as "normalcdf(445, 1E99, 454, 6)" with nothing labelled. Readers need the distribution, its mean and standard deviation, and the boundary identified. Unlabelled syntax, or a calculator output with no conditions or conclusion in context, loses credit even when the number is right.

  • Spending three or four minutes computing an answer to a conceptual question. If the stem asks what happens to the margin of error when n goes from 100 to 400, the answer is that it halves. A full interval calculation can't make that answer more correct.

Checklist

  • I can sort a multiple-choice stem as conceptual or computational in about 10 seconds, before I touch the calculator.
  • I can name the exact command for a normal area, a normal cutoff, a binomial "exactly", a binomial "at most" and a binomial "at least".
  • I can turn words into calculator inputs without an error: "at least k" for a binomial is 1 − binomcdf(n, p, k − 1), "more than k" for a binomial is 1 − binomcdf(n, p, k), and "more than x" for a normal is normalcdf(x, 1E99, μ, σ).
  • I always report Sx, not σx, for sample data, and I enter standard deviations, not variances.
  • I can choose between a z and a t procedure, and I leave Pooled set to No for a two-sample t.
  • I can estimate a normal probability with the 68–95–99.7 rule well enough to eliminate wrong choices.
  • On free response, I name the distribution or procedure and its parameters, check conditions in writing, and finish with a conclusion in context.
  • I can answer "does this interval support the claim?" by comparing the claimed value to the interval, without running a second test.

More AP Statistics technique